Alphabeta Math
LemmaStatement: AI-adaptedProof: AI-generatedPipeline-generatedprecheck passjudge pass (gpt-6-luna)
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

A principal-angle bound encloses the largest selected generalized Ritz value across a spectral gap

Statement

Let A=A∗∈Cn×n, B=B∗>0, and λ1≤⋯≤λn be the generalized eigenvalues of (A,B). Fix 1≤q<n with the strict gap λq<λq+1 and let U=span⁡{u1,…,uq} be the corresponding sum of the first q eigenspaces, with (ui) B-orthonormal. For a q-dimensional trial subspace S with B-orthonormal basis Q, define PUx=∑i=1qui(ui∗Bx),sin⁡ϑ=sup⁡x∈S∥x∥B=1∥(I−PU)x∥B. Let θ1(S)≤⋯≤θq(S) be the generalized Ritz values from Generalized Ritz values obey min-max bounds and are monotone under trial-space inclusion. Then λj≤θj(S)(1≤j≤q),θq(S)≤λq+(λn−λq)sin⁡2ϑ. The strict gap identifies U uniquely; the estimate is a subspace enclosure, not an iteration convergence rate.

Facts & Assumptions

Given: A Hermitian-definite pair (A,B), its ordered generalized eigenvalues and B-orthonormal eigenbasis, the strict gap at q, and a q-dimensional B-orthonormal trial basis Q spanning S.

[F1]

The B-inner product, B-norm, and dual residual norm: ⟨x,y⟩B=y∗Bx and ∥x∥B2=x∗Bx.

[F2]

A Hermitian-definite pencil has a real spectrum and a B-orthonormal eigenbasis: A and B have a B-orthonormal generalized eigenbasis with real eigenvalues.

[L1]

Generalized Ritz values obey min-max bounds and are monotone under trial-space inclusion: each selected Ritz value is at least the corresponding ambient generalized eigenvalue, and Ritz values decrease when the trial subspace is enlarged.

Proof

technique · direct
1.1givenF1F2algebra

By [F2], write a B-unit vector x∈S as x=u+e with u=PUx∈U and e=(I−PU)x∈U⊥B. In the B-orthonormal generalized eigenbasis, if x=∑iaiui, then x∗Bx=∑i∣ai∣2 and x∗Ax=∑iλi∣ai∣2. Thus u∗Au≤λqu∗Bu and e∗Ae≤λne∗Be; the cross terms vanish since ui∗Auj=λjui∗Buj=0 for i≤q<j. Therefore x∗Ax/x∗Bx≤λq∥u∥B2+λn∥e∥B2=λq+(λn−λq)∥e∥B2, which is at most the stated angle bound.

2.1givenF2L1step 1.1algebra∎

Diagonalizing the finite Hermitian compression shows that θq(S) is the maximum Rayleigh quotient on S, so step 1.1 gives its upper bound. The lower bounds λj≤θj(S) are [L1]. The strict gap isolates the first q eigenspaces, so their sum is unique; no step asserts an iteration rate.

Depends on

Used by

Dependency tree · two levels

14 results within two dependency steps of this one, each drawn at its shortest distance from it. An arrow runs from a result to what uses it, so the chart reads left to right and ends at this result, which carries a heavier outline. Every node is a link to that result. Click elsewhere on the chart to enlarge it.

Sources